1. Introduction
It has been shown that the use of high horizontal resolution (less than 75-km grid spacing) improves climate simulation in many ways. Phenomena operating at spatial scales too small for traditional general circulation models (GCMs) become better resolved with finer grid spacing. These include tropical cyclones (e.g., Zhao et al. 2009; Manganello et al. 2012; Wehner et al. 2014; Zarzycki and Jablonowski 2014) and frontal zones (Ohfuchi et al. 2004). The diurnal cycle of precipitation is better simulated with increased resolution, particularly in models that are convection permitting (Dirmeyer et al. 2012). Additionally, increased resolution provides a more accurate topographical boundary condition. Better orographic representation has been shown to improve precipitation patterns (Gent et al. 2010; Boyle and Klein 2010) and midlatitude blocking events (Jung et al. 2012).
However, integrating for long periods of time (multiple decades) at these grid spacings is not standard, even at large national centers dedicated to atmospheric modeling. Additionally, multimember ensemble simulations provide insight into sources of uncertainty within a model simulation (Rougier et al. 2009; Flato et al. 2014). This insight cannot be gained from a single model run at fine horizontal resolution, which may make decreasing grid spacing and exhausting available computational resources on a single simulation a less-than-ideal option.
Variable-resolution general circulation models (VRGCMs) hold the potential to significantly improve regional climate simulations. VRGCMs employ either grid stretching (e.g., Fox-Rabinovitz et al. 2006, and references therein; Tomita 2008) or grid refinement (e.g., Ringler et al. 2008; Jablonowski et al. 2009; Walko and Avissar 2011; Skamarock et al. 2012; Rauscher et al. 2013; Harris and Lin 2013, 2014; Zarzycki et al. 2014a,b) to only simulate a portion of the global domain at high resolution.
The bulk of recent VRGCM development has centered around grid refinement techniques, since stretched grid models can only have a singular high-resolution region and stretching results in coarser grid spacing opposite of the high-resolution region. This causes adverse effects at higher stretching factors that degrade the solution quality in the far-field region (Caian and Geleyn 1997; Lorant and Royer 2001).
VRGCMs focus available computing resources in high-resolution areas while lessening the cost required to simulate the global circulation over the remainder of the (coarser) domain. This decrease in computational burden, while still allowing regionally high resolutions within a global framework, may support improvements such as longer simulations and the addition of members within ensembles. For example, if a variable-resolution (VR) setup only requires one-tenth of the computing wall clock time a globally uniform grid necessitates, 10 ensemble members can be run in lieu of one high-resolution model simulation assuming identical resources.
Therefore, VRGCMs can be considered a bridge between traditional, more computationally expensive GCMs with uniform grid spacing and limited area models (LAMs), which require forcing from lateral boundary conditions. These boundary conditions are generally derived from a different driver model or dataset and may be poorly interpolated, mathematically ill posed, or physically inconsistent (Warner et al. 1997; McDonald 2003; Laprise et al. 2008; Mesinger and Veljovic 2013). These issues can be mitigated using a properly formulated global VRGCM.
Recently, a VR option has been implemented within the Community Atmosphere Model’s (CAM) spectral element (SE) dynamical core (Neale et al. 2010b). CAM is jointly developed by the National Center for Atmospheric Research (NCAR) and various Department of Energy (DOE) laboratories. Variable-resolution CAM-SE has shown promise in allowing for high-resolution simulation of tropical cyclones (Zarzycki et al. 2014a; Zarzycki and Jablonowski 2014). In addition, related aquaplanet simulations with CAM-SE exemplify that refined nests can accurately reproduce the regional climatology of a uniform high-resolution run at a fraction of the computational expense (Zarzycki et al. 2014b).
In this paper, we increase the complexity of our previous variable-resolution CAM-SE assessments and report on two 23-yr simulations that follow the Atmospheric Model Intercomparison Project (AMIP) protocol (Gates 1992). CAM is forced with prescribed sea surface temperatures (SSTs) and ice coverage in an attempt to recreate the observed climatology of the last three decades. One simulation utilizes a globally uniform 1° (~111 km) grid, analogous to resolutions used in recent climate assessments. The other simulation uses the same grid with an embedded 0.25° (~28 km) nest over the Atlantic Ocean. The reason for this refinement was to investigate the performance of CAM-SE at simulating tropical cyclones using multiresolution nests. Those results are contained within a companion paper (Zarzycki and Jablonowski 2014).
This paper discusses the long-term climatological state of both simulations. We pay particular attention to changes in regional climate arising from the addition of a high-resolution nest. We also investigate the scale sensitivity of the CAM, version 5 (CAM5), physical parameterization package and whether the addition of refinement degrades large-scale circulation patterns or climate statistics. The paper is structured as follows. In section 2 we briefly discuss CAM-SE and the special considerations for variable resolution, including the development of a multiresolution topographical dataset. Section 3 examines the climatological averages and the spatial effects of an embedded nest on long-term means. Section 4 discusses equatorial waves, particularly African easterly waves, and the impact of refinement on their representation in the model. Section 5 outlines a few examples of regional climate improvements that arise from the use of variable resolution. Section 6 summarizes the results and discusses the potential implications of this work as well as future research directions.
2. Model description and experimental setup
a. CAM-SE
The SE dynamical core in CAM is based upon a continuous Galerkin spectral finite-element method applied on a cubed-sphere grid (Taylor et al. 1997; Thomas and Loft 2005; Taylor and Fournier 2010; Dennis et al. 2012). The use of the quasi-uniform, cubed-sphere mesh eliminates problems arising from converging meridians on standard latitude–longitude grids. CAM-SE locally conserves mass and tracer mass to machine precision, as well as moist total energy to the level of time truncation error in the absence of dissipative processes (Taylor 2011). The primitive equations governing atmospheric motion are solved locally on individual elements, reducing the amount of interprocessor communication required with other numerical schemes. This gives CAM-SE attractive scaling properties; the model has been shown to scale nearly linearly to hundreds of thousands of cores (Dennis et al. 2012; Evans et al. 2013). These characteristics make CAM-SE a compelling option for future high-resolution climate simulations on massively parallel systems.
Because the discretization is localized on individual elements, variable resolution can be introduced through refined meshes provided the elements tiling the sphere are conforming quadrilaterals. This setup allows VR grids to maintain the key conservation and scalability aspects that make CAM-SE a desirable model choice for climate simulations.
b. Experimental setup
This study utilizes version 1.1.17 of the Community Earth System Model (CESM). CESM is a coupled climate system model combining CAM with other model components such as land and ocean.
We run two simulations with two different CAM grids. One is a globally uniform 1° (~111 km) CAM-SE grid. We refer to this as the “coarse” simulation. This is the default model grid for CAM simulations as of version 5.3. The other is a refined mesh that uses the same 1° grid with a patch of 0.25° (~28 km) refinement embedded over the Atlantic Ocean. A small transition band of 0.5° (~55 km) grid spacing separates the inner nest from the coarser, background grid. This is referred to as the “VR” mesh. Both grids can be seen in Fig. 1. The irregular shape of the high-resolution patch was determined by historical tropical cyclone activity in the North Atlantic Ocean basin. The grid generation procedure and refinement structure is detailed in Zarzycki et al. (2014b).
The two meshes used for this study are (a) a uniform 1° resolution mesh and (b) a VR mesh that ranges from 1° to 0.25°. Note that each element shown contains additional 3 × 3 collocation cells.
Citation: Journal of Climate 28, 7; 10.1175/JCLI-D-14-00599.1
There are only a few modifications required to utilize CAM-SE in conjunction with VR grids. CAM-SE applies explicit fourth-order hyperviscosity both for numerical stability and to simulate a realistic kinetic energy spectrum (Dennis et al. 2012). This fourth-order hyperdiffusion must be scaled with resolution such that the proper damping is applied at the coarse grid scales without harming dynamically resolved features in the high-resolution nest. The fourth-order diffusion coefficient is scaled according to the grid size of each element with higher (lower) values in larger (smaller) grid boxes. Further discussion of the scalar hyperviscosity used in this study can be found in Levy et al. (2013), Zarzycki et al. (2014a,b), and Guba et al. (2014).
The default polynomial degree in CAM-SE is chosen to be three, which is the operational default in CAM. This selection leads to fourth-order spatial accuracy. We utilize a finite difference approach in the vertical with a hybrid sigma-pressure coordinate as well as a Runge–Kutta time discretization. All simulations use 30 vertical levels with a model top of approximately 2 hPa. The model is therefore only refined in the horizontal direction. The dynamical time step 
We utilize the CAM5 subgrid physics package (Neale et al. 2010b) to parameterize processes not explicitly resolved by the dynamical core. CAM5 is the newest set of physical parameterizations available within CESM and has been shown to be the superior choice within CAM for VR simulations because of improved scaling of cloud fraction and precipitation at multiple resolutions when compared to prior versions (Zarzycki et al. 2014b). To minimize computational cost incurred by the addition of the new 3-moment interactive modal aerosol model (MAM; Liu et al. 2012) in CAM5, we utilize a prescribed aerosol configuration similar to the bulk aerosol model (BAM; Kiehl et al. 2000) used in previous versions of CAM. A more detailed description regarding the prescribed BAM aerosol setup in CAM5 can be found in Bacmeister et al. (2014).
The parameterizations are identical in every grid cell, regardless of resolution. While selectively tuning parameterizations to a specific grid spacing may be a target for future research, Harris and Lin (2014) showed that any improvement in model solution through differential tuning within VR simulations was much smaller than improvements achieved through better representation of orography and other dynamical features within a fine nest. Cold ice and rainwater autoconversion coefficients were set to match the defaults for high-resolution (0.25°) CAM simulations using the FV dynamical core. All other physical parameterization tuning parameters are non-resolution-specific CAM defaults that are derived from 1° finite volume CAM (CAM-FV) simulations. These tuning parameters are identical to recent simulations using CAM-FV at 0.25° resolution (Bacmeister et al. 2014; Wehner et al. 2014), and the adoption of CAM-FV tuning parameters for CAM-SE has been used with success in past studies (Evans et al. 2013; Zarzycki et al. 2014b).
The simulations follow the AMIP protocols first outlined in Gates (1992). SSTs and ice coverage are applied through the monthly 1° Hadley Centre Sea Ice and Sea Surface Temperature dataset (HadISST; Hurrell et al. 2008). Greenhouse gas concentrations and aerosol climatology are prescribed based on past observations. The atmospheric grid is coupled to ocean/ice and land models through the CPL7 trigrid coupler (Craig et al. 2012), which allows fluxes passed between the atmosphere and other model components to be conservatively remapped to the different grids. We utilize the Community Land Model (CLM), version 4.0, run on a 0.9° × 1.25° latitude–longitude grid. The land model is not prescribed and freely adjusts with the climate system.
Both simulations are initialized in September of 1979. While the initial conditions for the atmosphere and land are regridded from previously spun-up cases utilizing CESM’s AMIP configuration, the first four months are discarded to allow the model to adjust to potential small imbalances arising from grid resolution and topography. Both runs continued through the middle of 2003, although only fully simulated calendar years (1980–2002) are analyzed. The VR simulation was completed on NCAR’s Bluefire machine in late 2012 and averaged ~0.42 simulated years per day (SYPD) on 384 processors. The globally uniform 1° simulation was run on the Agri computing cluster at the University of California, Davis, in mid-2013 with a model throughput of about 2.5 SYPD on 384 processors. We note that a direct scaling analysis between the two runs is not possible because of the different hardware architectures of the two systems.
c. Generation of topographical dataset for VR
One particular challenge that arises from the use of VRGCMs is the representation of topography. CAM-SE, which uses terrain-following coordinates, requires smoother topography at coarser resolutions to maintain numerical stability and prevent numerical artifacts such as Gibbs ringing. In addition, topography that is too rough has been shown to produce spurious vertical velocities within CAM-SE (Evans et al. 2013).
To generate surface topography data for variable-resolution CAM-SE, the CAM-FV default topography at 0.23° × 0.31° is regridded to the unstructured CAM-SE grid using bilinear interpolation. This is the highest-resolution dataset packaged with CAM. Cursory tests showed that there is no appreciable difference between using bilinear interpolation and an alternative high-order remapping scheme. Interpolating from 0.23° × 0.31° data is an acceptable option because CAM-SE requires slightly smoother topography fields than CAM-FV (Evans et al. 2013).
The VR topography is smoothed for 32 iterations with o equal to 2 and c equal to 120 s. The value of 
A comparison of the topography at different grid spacings for the two simulations is shown in Fig. 2. More finescale structure in surface geopotential is apparent in the high-resolution nest of the VR simulation (Fig. 2a) because of the scaling of 
Surface geopotential of the topography over the location of the high-resolution nest (North Atlantic) in the (a) VR simulation and (b) uniform 1° simulation. In (a), the innermost red contour encompasses the 0.25° grid spacing, the outermost contour bounds the 0.5° transition region, and 1° grid spacing lies outside the outermost contour.
Citation: Journal of Climate 28, 7; 10.1175/JCLI-D-14-00599.1
These parameters produce a slightly smoother result in the high-resolution region than the default CAM-SE topography datasets supplied with CESM at 0.25° resolution. However, default CAM-SE at 0.25° resolution applies a 2.52 times stronger divergent component of the explicit diffusion to allow for rougher topography (P. Lauritzen 2014, personal communication). Recent work has indicated that features such as tropical cyclones in climate models may be significantly affected by modifications of the explicit dissipation (Zhao et al. 2012) and, in particular, divergence damping. Therefore, we have opted to use smoother topography to avoid this increased diffusion of the divergent motion in an attempt to use as consistent a diffusion formulation as possible.
Last, values for subgrid variability of topography are recomputed for the smoothed data.1 These are used in the parameterization of turbulent mountain stress, subgrid orographic drag, and momentum flux deposition due to gravity waves (Lauritzen et al. 2012).
d. Observational datasets
As a reference baseline to compare the model solutions, we use a variety of observational and reanalysis products. These are shown in Table 1. We acknowledge that caution must be exercised when using reanalysis products (e.g., Bosilovich et al. 2011) as “truth.” However, in this case we are less concerned with the accuracy of these products, but rather, using them as realistic proxies constrained by observations that allow for a comparison of the model results. The main uses of these datasets are for normalization and subjective discussion of relative differences between the two model simulations.
Variables used to evaluate CAM model performance for both grids and the corresponding observational (ISCCP, TRMM) or reanalysis (MERRA, NCEP) dataset and their period used as a reference baseline.
3. Climatological averages
To assess whether or not the refinement has any significant impact on the model climatology, we investigate four spatial regions. These areas are plotted in Fig. 3. The first area is simply the entire global domain. We also subselect regions of equal area and latitude over both the North Atlantic (red, diagonal hatch) and North Pacific Oceans (blue, stippled). These areas represent locations where the grid spacing is different between the two simulations (Atlantic) and where it is the same (Pacific). For topographically modified flow, we also look at a fourth region centered over central and northern South America (green, crossed hatch) that contains the largest differences in orography between the two models (as seen in Fig. 2).
Averaging regions used in this study.
Citation: Journal of Climate 28, 7; 10.1175/JCLI-D-14-00599.1
a. Global averages
Global annual averages (for the full 23-yr simulation period) of select parameters for both the uniform 1° and VR simulations are listed in Table 2. In addition to the averages, the absolute difference between the two model runs (VR minus uniform 1°) as well as the percent difference (normalized to the 1°) are also listed. Note that, although dry mass is conserved in CAM-SE (Taylor 2011), global surface pressure (PS), or atmospheric mass, is slightly different between the two simulations because of differing model initial conditions and slightly more water loading in the VR simulation. Diagnostics indicate that dry mass in each simulation is conserved to machine truncation.
Global statistics averaged over the 23-yr simulation period for various climate metrics in both the uniform 1° and the VR simulation. VAR is the variable field name abbreviation, ΔVAR is the VR minus 1° difference, and ΔVAR (%) is the percentage difference normalized to the 1° run.
The simulated global average total precipitation (PRECT) rate of 3.11 mm day−1 in both simulations is in good agreement with previously published CAM5-FV results of 3.04–3.18 mm day−1 (Bacmeister et al. 2014). Modeled total precipitable water and cloud fraction are well within the CMIP3 and CMIP5 multimodel ranges (approximately 21–27 kg m−2 and 55%–75% respectively) reported by Jiang et al. (2012). In addition, the CAM global mean surface temperature for both cases is well-matched to observational estimates of approximately 14°C (Jones et al. 1999; Hansen et al. 2010).
All parameters other than the convective (PRECC) and large-scale precipitation rates (PRECL) do not differ between the simulations by more than 1.3%, with all analyzed variables other than total cloud fraction and surface sensible heat flux having relative differences of 0.1% or less. It is clear that the addition of the high-resolution nest in this study contributes very little in the way of significant changes in model averages at the global level.
Averages for just the Atlantic basin are shown in Table 3. The largest discrepancy between the two models runs is in the partitioning between convective and large-scale (stratiform) precipitation. The resolution-dependence of both components of the parameterized precipitation is a known behavior of CAM (e.g., Duffy et al. 2003; Williamson 2008; Boyle and Klein 2010; Li et al. 2011; O’Brien et al. 2013; Zarzycki et al. 2014b). As refinement is introduced, convective precipitation decreases by 11.2% while large-scale precipitation correspondingly increases by 79.9%. The strength and frequency of resolved dynamical updrafts increase with resolution, leading to increased activation of the large-scale microphysics routine, therefore resulting in more precipitation from that model component at finer grid spacing.
The difference in total precipitation (the sum of both the convective and large-scale precipitation components) is significantly smaller than either of the components when they are considered separately. In the Atlantic region, the additional precipitation from the large-scale routine in the VR simulation is largely offset by a decrease in parameterized convective precipitation. The total precipitation is slightly higher in the fine-grid simulation, which is in agreement with previous simulations investigating the scale-sensitivity of precipitation in CAM (e.g., Williamson 2008; Rauscher et al. 2013; Bacmeister et al. 2014; Wehner et al. 2014; Zarzycki et al. 2014b).
The magnitudes of both cloud fraction and surface sensible heat flux also show slight (less than 7%) increases when refinement is introduced to the Atlantic basin. The two significant contributors to the sensible heat flux formulation in CAM are surface wind stress and the temperature gradient between the sea surface and lowest model level. The sensible heat flux increase may be due to an increase in the frequency of extreme low-level wind speeds within the high-resolution nest that arise from finer resolution (not shown). This increases the stress on the ocean surface and increases the heat transfer to the lowest levels of the atmosphere. Both simulations utilize the same SST data and have highly similar 850-hPa temperature (T850) and 2-m reference height temperature (TREFHT) climatology, implying that the temperature gradient between the ocean surface and low atmosphere is not the driver. Interestingly, the latent heat flux also increases, but much less than the sensible heat flux. This may be due to the additional nonlinearity of moisture introduced in the formulation of the latent heat flux. A study more thoroughly investigating the model resolution sensitivity of the boundary layer and surface flux schemes is a target for future research.
Statistics for the Pacific basin are shown in Table 4. Since both simulations have identical grid spacing over this region, significant differences would be the result of potential upstream or downstream effects of the Atlantic refinement. All metrics exhibit a 2.6% or less difference in climatology between the simulations. This near-identical match of the climatology supports the conclusion that model behavior at the 1° grid spacing in the variable resolution is well matched to the corresponding globally uniform run.
b. Spatial differences
The global, annually averaged plot of the 200-hPa zonal wind (U200) is shown in Fig. 4. Figures 4a and 4b show the results from the two model runs and are virtually indistinguishable from one another, particularly over the refinement region outlined in black (Fig. 4a). The absolute difference between the two simulations is plotted in Fig. 4c, which further highlights the similarity between the simulations. The largest differences are in the midlatitude eastern North Pacific, but the maximum discrepancy between the two runs at any spatial location is less than 5%. The U200 NCEP reanalysis product is shown in Fig. 4d. Both model solutions appear to overestimate the midlatitude jet stream in the Southern Hemisphere. The differences between either model simulation and NCEP are much larger than the differences between the individual model simulations themselves.
Plot of annually averaged 200-hPa zonal wind (U200, m s−1) for the (a) VR and (b) uniform 1° simulations. Also shown are (c) the difference between (a) and (b) and (d) U200 from the NCEP reanalysis.
Citation: Journal of Climate 28, 7; 10.1175/JCLI-D-14-00599.1
To quantify this, we calculate the global root-mean-square error (RMSE). The RMSE for U200 between both CAM-SE configurations in Figs. 4a and 4b is 0.77 m s−1. In contrast, it is 2.34 and 2.32 m s−1 when the NCEP reanalysis is compared to the VR and uniform 1° CAM-SE simulations, respectively. This confirms that the differences between simulations using the two grids are extremely small relative to their discrepancy from the reanalysis data. The RMSE for either simulation compared to NCEP is also slightly smaller than the 2.87 m s−1 RMSE calculated in Evans et al. (2013), who used the same uniform 1° CAM-SE grid with CAM, version 4 (CAM4; Neale et al. 2010a), physics.
A latitude–height cross section of the same time-mean zonal wind is plotted in Fig. 5. Here, the analysis is constrained to the high-resolution nest. The longitudinal subset used for averaging (80°–20°W) is restricted to the Atlantic (red) grid box from Fig. 3. Black lines in Figs. 5a and 5c denote the approximate latitudes of the innermost grid transition region (between 0.25° and 0.5°). Both Figs. 5a and 5b are highly similar. The uniform 1° model has a slightly stronger and higher jet stream core (peaking at approximately 225 hPa, 47°N), although as seen in Fig. 5c, this difference is less than 7% (region above 200 hPa). As in Fig. 4, the differences between the two model simulations are much smaller than the difference between either simulation and the NCEP dataset (Fig. 5d). While the model simulations are highly similar to one another, the refined simulation does appear to represent a marginal improvement over the unrefined simulation with a reduced bias in zonal wind speed relative to the NCEP climatology.
Pressure latitude cross section of annually and zonally averaged zonal wind for the (a) VR simulation, (b) uniform 1° simulation, and (d) NCEP. (c) The difference between the two model simulations. The zonal average is taken between 80° and 20°W, which corresponds to the longitude bounds of the Atlantic refinement in Fig. 3. Black lines denote the latitude bounds of the averaging area (red box in Fig. 3).
Citation: Journal of Climate 28, 7; 10.1175/JCLI-D-14-00599.1
The spatial distributions of annually averaged column total precipitable water (TMQ) over the Atlantic for both models are plotted in Figs. 6a and 6b. Analysis of the global domain (not shown) shows similar results to Fig. 4 where the distribution of TMQ is highly correlated in the 1° cells of both runs. The absolute difference between the two simulations is plotted in Fig. 6c, with the average from the MERRA reanalysis over the 1980–2002 time period posted in Fig. 6d. TMQ is an interesting metric to examine because it is a rather smooth field; however, larger mountain ranges have significant regional impact on the spatial structure of the field in their immediate vicinity (Tuller 1968). Because we seek to emphasize the topographical enhancement provided by variable resolution, we have chosen to use MERRA because of its increased horizontal latitude–longitude resolution [0.5° × 0.66° (~55–70 km) as compared to NCEP’s T62 (~210 km) resolution]. Analysis using NCEP’s total precipitable water product showed that the resolution was too coarse to allow topography to impact regional TMQ structure in a meaningful way.
Plot of annually averaged TMQ (kg m−2) for the (a) VR and (b) uniform 1° simulations. Also shown are (c) the difference between (a) and (b) and (d) the TMQ from the MERRA dataset.
Citation: Journal of Climate 28, 7; 10.1175/JCLI-D-14-00599.1
Figures 6a and 6b show multiple key differences between the VR and uniform 1° TMQ fields. Structure associated with the Appalachian Mountain range (near 40°N, 80°W) is noticeable over eastern North America in Fig. 6a, along with a local minimum in TMQ over the island of Hispaniola (19°N, 72°W). In addition, the topographical signature of mountainous areas of Mexico, Central America, and northern South America are much more structured in the VR simulation, with the relative TMQ deficit being more significant than in the coarse, uniform 1° model run.
Unlike Figs. 4 and 5, the difference panel (Fig. 6c) shows significant spatial difference between the two models. The vast majority of the differences are constrained to regions where the topography has been more (less) smoothed in the uniform 1° (VR) simulation (Fig. 2). Conversely, there is very little notable difference across the different grid scales themselves. There is essentially no anomaly seen in Fig. 6c over the central Atlantic Ocean (the center of the refined nest). This is in contrast to grid-imprinting induced by highly scale-sensitive parameterization schemes [see Zarzycki et al. (2014b), their Fig. 5, for example of grid imprinting in CAM4 cloud fraction]. This result implies that the majority of the climatological impact of refinement on average TMQ is not directly due to the grid spacing but indirectly due to the rougher topography allowed by the use of variable resolution. The lack of observable difference over the Atlantic Ocean and statistics shown in Table 3 show that localized refinement does not impact the large-scale TMQ distribution. It is also clear that the MERRA product (Fig. 6d) much more closely resembles the VR simulation (Fig. 6a), especially in areas of the refined patch where the topographical representation is improved (Central America and northern South America). This implies that the refined resolution produces more topographically realistic flow.
We show similar analyses for total cloud fraction, total precipitation rate, and outgoing longwave radiation (OLR) in Fig. 7. All fields agree with the conclusions from investigating the TMQ climatology. The largest discrepancy resulting from the addition of the refined patch appears to be constrained in areas where the topography is better represented in the VR simulation. There are almost no distinguishable artifacts in the long-term means that appear as a result of refinement, either as artifacts in/near transition regions due to the numerical discretization or as an induced climate bias appearing over the central Atlantic due to the physical parameterizations behaving significantly differently in the refined nest. The fact that the spatial patterns match well is a positive result for the implementation of variable-resolution CAM-SE as a tool for regional climate studies, but it also confirms the results of Bacmeister et al. (2014), who concluded that substantial climate biases at large scales will likely not be improved through merely increasing resolution in CAM5.
Plots of annually averaged (a)–(c) vertically integrated cloud fraction, (d)–(f) total precipitation rate, and (g)–(i) OLR for the (left) VR, (middle) uniform 1° simulations, (right) the difference between the two.
Citation: Journal of Climate 28, 7; 10.1175/JCLI-D-14-00599.1
Note that there is a degree of grid imprinting that appears at the southern interface between the intermediate 0.5° grid spacing and the global 1° grid (solid line stretching from Peru to Brazil over the Amazon). This is primarily evident in the precipitation field (Fig. 7), but also distinguishable in both OLR and cloud fraction. For example, the VR simulation produces slightly less precipitation to the north and slightly more precipitation to the south of this grid transition region than the global 1° simulation.
This imprinting was not evident until the run was completed and long-term means were calculated since it is not overtly discernible in instantaneous output. A bug in the Laplacian operator within the dynamical core was rectified between CESM version 1.1.17 (the developmental version used in this study) and the release version 1.2.0. This issue was only noticeable in highly deformed elements (such as the refined region’s southern transition). To verify this issue was corrected, we completed a 2-yr simulation using the release version of CESM (1.2.0). All user-defined model settings such as time step, model tunings, and topography are identical to the full (1.1.17) simulation. A zoom of the 1999–2000 average precipitation is shown in Fig. 8. The artifacts that appear in the time mean of the 1.1.17 simulations are eliminated in the newer version of CESM 1.2.0. Future simulations will utilize CESM version 1.2.0 or higher to ensure these fixes are used going forward.
Comparison of annually averaged precipitation rate for 1999–2000 between VR simulation using (a) the CESM 1.1.17 developmental release and (b) CESM 1.2.0.
Citation: Journal of Climate 28, 7; 10.1175/JCLI-D-14-00599.1
c. Taylor statistics
Taylor diagrams depict how well matched a spatial pattern produced by a model simulation is to observations (Taylor 2001). The two simulations are compared to the observational datasets in Table 1. For the majority of this analysis we are primarily concerned with the relative difference between the two models and, therefore, whether or not the climatological skill of the model is significantly altered by the addition of refinement. A thorough analysis understanding why certain parameters are modeled with their particular degree of skill in CAM itself is beyond the scope of this paper.
Taylor statistics for 23-yr-mean global-mean quantities for sea level pressure (PSL), total cloud fraction (CLDTOT), total precipitable water (TMQ), total precipitation rate (PRECT), 200-hPa zonal wind (U200), 850-hPa zonal wind (U850), 600-hPa relative humidity (RH600), and 500-hPa temperature (T500) are shown in Fig. 9. The absolute distance from the origin (lower left) represents the magnitude of the spatial variability within the domain (as measured by normalized standard deviation) while the spatial correlation is plotted as the radial angle between the model marker and the origin. Perfect agreement between model and observations would result in a marker being plotted at 1.0 correlation and at the “REF” location on the x axis. A comprehensive discussion of Taylor diagram generation can be found in Taylor (2001). Filled red dots indicate the VR simulation skill while blue crossed circles are from the uniform 1° simulation. Note that TRMM data are unavailable poleward of 50° latitude (Huffman et al. 2007), so high-latitude regions were not used in the calculations for precipitation.
Taylor diagram for globally and annually averaged climate statistics. Blue crossed circles represent the uniform 1° simulation, while the filled red circles represent the VR run. Datasets used as observations are listed in Table 1. See text for description of the diagram and explanation of acronyms.
Citation: Journal of Climate 28, 7; 10.1175/JCLI-D-14-00599.1
As in Table 2, the model shows good agreement between the two simulations in terms of long-term climatology. The fact that the majority of the corresponding variable score pairs either overlay one another or are very close show that the spatial distribution and magnitude of these variables are essentially identical to one another on both grids. Skill scores for both simulations are almost identical to those reported in Bacmeister et al. (2014; their Figs. 2 and 3), who assessed globally uniform CAM simulations at ~1° and ~0.25° grid spacings, establishing that the VR setup produces a climatology that mirrors that of previously published CAM simulations.
Figure 10 shows seasonal [December–February (DJF) and June–August (JJA)] statistics broken out into the three shaded regions from Fig. 3. There is more separation in some of the point pairs, indicating potentially different solutions at the regional scale, although the smaller domain and seasonal breakdown results in fewer data points in both space and time. Figures 10a and 10b show Taylor statistics in only the North Atlantic region marked in Fig. 3. Refinement appears to have little quantifiable effect on TMQ, PRECT, U850, and T500 when the analysis is restricted to the high-resolution part of the Atlantic basin. PSL is slightly harmed in DJF while improved in JJA when refinement is added. Little change is noted in CLDTOT during winter months. The uniform 1° simulation has higher CLDTOT skill in summer, although this quantity is poorly simulated in both model runs during this season. U850 and RH600 improve slightly in both seasons with the addition of resolution.
As in Fig. 9, but broken down by season and region: (left) DJF and (right) JJA averages. Regions are defined in Fig. 3. (a),(b) North Atlantic; (c),(d) North Pacific; and (e),(f) Central/South America.
Citation: Journal of Climate 28, 7; 10.1175/JCLI-D-14-00599.1
Figures 10c and 10d show the same analysis except for the North Pacific. In these plots, particularly the summer, corresponding pairs are not separated by significant distance, implying the model solution in both runs is highly similar. Interestingly, PSL is significantly degraded in the winter in the VR simulation, while PRECT has a slightly lower correlation, but improved variability. It is unclear if these represent upscale effects, especially since wintertime differences in the North Atlantic were relatively small (Fig. 10a) and very high agreement is shown between the two model simulations during the Pacific summer (Fig. 10d). It is possible some of this difference may stem from slightly different Himalayan topography that was alluded to in section 2c. These small differences may have a minor impact on certain patterns in the North Pacific.
The most interesting analysis is in Figs. 10e and 10f, which highlight Taylor statistics over the Central American region (Fig. 3, green crosshatch). This is the region where the largest differences between the simulations were seen and may be attributable to the improved topographical representation in the VR simulation. Winter-time PRECT and TMQ are dramatically improved in the VR simulation, with a smaller improvement in CLDTOT. All three of these features would be expected to have a strong relationship with the model topography since topography exerts direct control on vertical motion. In the summer, TMQ is similarly improved with resolution, while PRECT shows a more modest improvement. CLDTOT correlation decreases slightly but the high bias in spatial variability is slightly smaller than in the uniform 1° run.
PSL shows a decrease in skill as measured by both quantities in both seasons. This result is similar to that seen in Bacmeister et al. (2014), who found a decrease in performance with higher resolution in aspects of CAM5, such as surface pressure, that may be related to the model’s turbulent mountain stress (TMS). TMS in CAM5 is intended to add surface stress because of unresolved subgrid orography (Richter et al. 2010). Bacmeister et al. (2014) postulate that the TMS tunings (which are tuned for ~1° simulations) may result in negative effects with increased resolution. RH600 also shows a slight decrease in both skill measurements with resolution during the summer, although it is unclear if this is related to the TMS parameterization or another mechanism that influences midlevel moisture or temperature. All other quantities in both seasons appear to have unremarkable differences between the two simulations, even in this smaller region with highly disparate topography.
4. Equatorial waves
One common method for detecting atmospheric waves is to generate wavenumber–frequency diagrams using the spectral decomposition methodology of Wheeler and Kiladis (1999). Figure 11 shows the wavenumber–frequency diagrams for both the uniform 1° simulation (left) and the VR simulation (right). The spectral analysis here uses 3-hourly total precipitation rates averaged between ±10°, which produce similar results to the more commonly used outgoing longwave radiation (not shown). Both the antisymmetric (Figs. 11a,b) and symmetric components (Figs. 11c,d) of the power spectra are normalized by the background component to show the most active waves.
Wavenumber–frequency diagrams of total precipitation rate averaged between 10°N and 10°S. (a),(b) Normalized antisymmetric and (c),(d) normalized symmetric components of the logarithm of the power are shown for the (left) coarse (uniform 1°) and (right) VR simulations. Dispersion curves from linear shallow-water theory for a zero wind basic state with equivalent depths h = 12, 25, and 50 m are overlaid as in Wheeler and Kiladis (1999). IG, ER, EIG, and Kelvin waves are marked with their meridional mode numbers n.
Citation: Journal of Climate 28, 7; 10.1175/JCLI-D-14-00599.1
The diagrams are generated using 96-day segments with 60 days of overlap. Precipitation from 1992 to 2001 is used. Following Wheeler and Kiladis (1999), dispersion curves at equivalent depths of h = 12, 25, and 50 m for n = 0 eastward inertio-gravity waves (n = 0 EIG), n = 2 inertio-gravity waves (n = 2 IG), and mixed Rossby-gravity waves (MRG) are shown in the antisymmetric panels (top) with the same dispersion curves for n = 1 equatorial Rossby waves (n = 1 ER), n = 1 inertio-gravity waves (n = 1 IG), and Kelvin waves shown on the symmetric panel (bottom).
Figures 11a and 11b show low wave activity in the antisymmetric power spectra. The highest peak occurs in westward-propagating waves with frequencies between 0.15 and 0.25 cycles per day. This peak is associated with MRG waves. In Figs. 11c and 11d, low-frequency Kelvin waves are prominent, with the most variance located between the 25- and 50-m equivalent depth curves at longer periods (lower frequency). The n = 1 ER waves are also present, albeit weaker, and there is no significant n = 1 IG wave activity.
These results are in good agreement with previous work using CAM-SE to perform AMIP simulations (Evans et al. 2013). Using CAM4 physics and a uniform 1° grid, their results also showed a robust Kelvin wave signature at lower wavenumbers that decreases with frequency. Furthermore, their ER waves were also simulated with reasonable skill, but the model failed to produce peaks corresponding to westward-propagating IG waves as well as MRG waves outside of the wavenumber band seen in this study. Additional comparison of wave features in the SE dynamical core using aquaplanet simulations can be found in Mishra et al. (2011).
Most relevant for this study, Fig. 11 indicates good agreement between the two simulations. The high-resolution patch covers approximately 35% of the equatorial region north of the equator. There does not appear to be spurious wave reflection or distortion induced by the grid transition region that would be denoted by the appearance of an anomalous power peak in either Figs. 11b or 11d. Additionally, the refinement does not adversely affect waves already resolved in the 1° simulation. Both simulations produce the same wave activity without any shift in phase speed or power. We note that, since the Atlantic Ocean is the climatologically least active basin with respect to equatorial wave activity (Kiladis et al. 2009) and that 0.25° grid spacing remains much coarser than the convective-permitting threshold, we do not expect to see significant increases in power within the VR simulation that might be discernible in simulations with very high-resolution around the entire equatorial band.
5. Regional climatology improvements
a. African easterly waves
African easterly waves (AEWs) are dynamical features originating over North Africa. These waves are related to the African easterly jet (AEJ) present in the midtroposphere, south of the Saharan desert. The source region for AEWs is between 32° and 15°E and centered at around 16°N (Burpee 1972; Reed et al. 1977). AEWs occur in the lower troposphere, near 700 hPa, during the summer months with a periodicity of about 3–5 days (Burpee 1972). The meridional wind has a maximum amplitude of 1–2 m s−1. The waves travel across the Atlantic Ocean, occasionally reaching the eastern Pacific Ocean, and play a key role in the generation of tropical cyclones in the tropical Atlantic (Frank 1970).
In the VR simulation, the area most strongly associated with AEW genesis straddles the transition region as seen in Fig. 1. Given this and the impact of AEWs on tropical cyclogenesis, they are an interesting case study for assessing the performance of the model from both a physical and dynamical standpoint. Figure 12 shows the vertical cross section of the zonal wind between 20°S and 40°N (averaged between 25°W and 15°E) for the 1° coarse simulation (left) and VR simulation (right). Winds are averaged over the Northern Hemisphere summer [June–September (JJAS)] seasons from 1990 to 2000. The jet associated with westerly monsoon winds is located at 10°N extending from the surface to around 850 hPa. The AEJ is centered around 15° and 650 hPa with the tropical easterly jet (TEJ) located around 6°N and 200 hPa. Note that the longitudinal span cuts directly through the transition region between the high-resolution nest and the background 1° grid. No material difference is seen between the two simulations. The strengths of both the TEJ and AEJ are essentially identical and the shape and location of these two features are the same in both simulations. Both jets span at least 10° in latitude and are well-resolved at the various grid spacings in the simulations. Their impressive similarity indicates that the resolved flow can transit through transition regions and adapt to multiple grid spacings in a physically consistent manner.
CAM-SE average zonal wind during JJAS for (left) the uniform 1° simulation and (right) the VR simulation. Data are averaged between 25°W and 15°E over the time period 1990–2000.
Citation: Journal of Climate 28, 7; 10.1175/JCLI-D-14-00599.1
AEWs are detected by analyzing the meridional wind speed at 700 hPa. The area of interest, 20°S–40°N and 50°W–40°E, covers West Africa and the eastern portion of the Atlantic Ocean. The meridional wind is averaged daily and bandpass filtered to isolate the activity with a frequency of 2–6 days during the JJAS summer months for each year. The variance is computed and then averaged over 20 years (1981–2000) to produce a measure of the typical AEW activity over West Africa.
Figure 13 shows the AEW activity in both model runs. In Fig. 13a, the solid black lines mark the outline of the transition region in the VR mesh. The overall structure is comparable with the uniform 1° results, and there is no discernible grid imprinting, even along the lines that indicate the region where the grid resolution changes. This again indicates the transition regions are being well-handled by the model and suggests that tropical cyclone simulations using a multiresolution grid within CAM-SE may not require that the high-resolution region extend over the entire continent of Africa to properly resolve wave precursors to cyclone genesis in the Atlantic Ocean.
Variance of 700-hPa meridional wind for (left) the uniform 1° simulation and (right) the VR simulation. Data are 2–6-day bandpass filtered and averaged over JJAS for the period 1981–2000.
Citation: Journal of Climate 28, 7; 10.1175/JCLI-D-14-00599.1
Additionally, the increased resolution appears to favor enhanced wave activity off the coast of Senegal (a regional centered on approximately 13°N, 20°W). This result is more in line with the multimodel results described in Skinner and Diffenbaugh (2013), which tend to show peak activity in similar regions. It is worth noting that when compared to the reanalysis data (not shown), the magnitudes of the CAM-SE wind variance are much higher (using either grid), suggesting that the model may produce unrealistically high AEW activity. However, wave activity in CAM-SE falls well within the envelope of models discussed in Skinner and Diffenbaugh (2013), implying CAM performs similarly to other comparable global models. The total spread in model results from Skinner and Diffenbaugh (2013) is an order of magnitude greater than the differences seen between the coarse uniform and VR simulations in Fig. 13, further showing that the model produces a highly similar solution with or without the presence of a high-resolution nest.
b. Precipitation extremes
Representation of extreme events lying in the tails of precipitation distributions continues to be problematic for the current generation of climate models, with models generally underrepresenting the frequency of these high-intensity events, especially in tropical locations (Mehran et al. 2014). The frequency of these precipitation events has been shown to increase with resolution in both earlier versions of CAM (Williamson 2008; Li et al. 2011) as well as CAM5 (Wehner et al. 2014; Zarzycki et al. 2014b). Finer resolution results in sharper gradients of moisture and temperature as well as stronger resolved updrafts. Figure 14 shows precipitation histograms for the Atlantic (top) and Pacific (bottom) statistical regions. Statistics are calculated using the 3-hourly total precipitation rate. Precipitation rates are first conservatively remapped to a uniform 2° grid based on the recommendations of Chen and Knutson (2008). Side-by-side panels show the same data, with the left panel only focusing on precipitation values less than 100 mm day−1.
Precipitation histogram representing fraction (logarithmic scale) of instances where 3-hourly instantaneous precipitation rates were in specific intensity bins for AMIP simulations. Statistics are averaged over (a),(b) North Atlantic (NATL) region and (c),(d) North Pacific (NPAC) from Fig. 3. The uniform 1° simulation (UNI) is plotted in red and the VR simulation in blue. Bin sizes are 1 mm day−1 on the left and 10 mm day−1 on the right.
Citation: Journal of Climate 28, 7; 10.1175/JCLI-D-14-00599.1
The 0.25° nest over the Atlantic produces an obvious divergence in precipitation frequency from the 1° grid spacing beyond 32 mm day−1, with the variable resolution simulation producing a much higher frequency of more intense events. Over the central Atlantic basin, the 1° simulation only produces a maximum precipitation rate of approximately 200 mm day−1, while the VR simulation produces events greater than 400 mm day−1, even after remapping. Figure 14a shows that, for precipitation rates less than approximately 32 mm day−1, the frequency of light precipitation events is higher in the coarse grid. While Table 3 describes a small (3%) increase in precipitation in the high-resolution nest compared to the 1° simulation, it is clear that the increase in high-intensity events is at least partially offset by a decrease in high-frequency, low-intensity “drizzle” events at the finer grid spacing. Figures 14c and 14d display nearly identical frequency profiles, which suggests that the dynamical behavior in the 1° portion of the VR grid is the same as in the globally uniform 1° simulation.
c. Mean precipitation climatology
While the direct simulation of transient small-scale features such as tropical cyclones, squall lines, and other extreme mesoscale features are popular targets for regional refinement, orographically influenced climatology also may be improved through the use of multiresolution grids.
In Figs. 15a–c we plot the mean climatology of total precipitation rates over Central America and the nearby bodies of water for both model simulations as well as observations. Model topography for the corresponding region is plotted in Figs. 15d and 15e. The National Geophysical Data Center (NGDC) 2-min (~3.5 km) Gridded Global Relief Dataset (ETOPO2v2) topography is shown in Fig. 15f.
Annual average total precipitation rate in the (a) VR and (b) uniform 1° simulations as well as (c) TRMM observations. Topography for the same regions is shown for (d),(e) both models as well as (f) the NGDC dataset. The transition boundary between 0.25° and 0.5° in the VR grid is highlighted in black and red in (a) and (d), respectively.
Citation: Journal of Climate 28, 7; 10.1175/JCLI-D-14-00599.1
The precipitation fields within the VR simulation (Fig. 15a) are noisy. However, more structure is apparent when compared to the unrefined 1° simulation (Fig. 15b). In particular, local maxima in precipitation are seen over various islands in the Caribbean Sea (blue and green contours around 20°N). This is also present in the TRMM observations (Fig. 15c). Additionally, the broad local minimum of precipitation over the Cordillera Isabelia mountain range in Honduras and Nicaragua (approximately 14°N, 88°W) is present in both the VR and TRMM data but not the 1° simulation. These results are similar in nature to the TMQ analysis in Fig. 6 and represent improved regional results in the VR simulation because of rougher topography.
There is a high bias in the local maximum in precipitation off the western coast of Colombia in both simulations. This anomaly also appears in previous 1° CAM4 AMIP simulations using both the SE and FV dynamical cores (Neale et al. 2013; Evans et al. 2013). This bias is greatly reduced in the VR simulation. We hypothesize that some of this improvement stems from the fact that less smoothing of the western slopes of the Andes is required in the VR simulation. Smoothing of orography in the 1° simulation has pushed the mountains 200–300 km into the Pacific Ocean (Fig. 15e). This leads to anomalously forced upslope flow over model grid boxes still masked as ocean cells, leading to dramatically increased precipitation. Neale et al. (2013) also showed that the bias (at 1° resolution) was somewhat reduced through coupling to an active ocean model.
The improved orography with the VR grid also simulates rain shadowing on the eastern side of the Andes Mountains (manifested as areas of suppressed precipitation oriented from north-northeast to south-southwest across northwestern South America) that is not seen in the uniform 1° simulation. While the 1° simulation has too much precipitation in many of these areas, the variable resolution is biased dry, however.
d. Diurnal cycle of precipitation
The annually averaged diurnal cycles of precipitation during the final 3 years of both simulations (2000–02) are plotted in Figs. 16a and 16b. Corresponding TRMM observations are shown in Fig. 16c. The hue (color) represents the local time of maximum precipitation occurrence while the saturation (intensity) denotes the magnitude of the diurnal cycle. In general, CAM produces a maximum in precipitation that is too early in the day across most land areas. This behavior is driven by premature convective initiation over land during warm-season months and is a common issue among current climate models (Dai 2006). This bias is especially prevalent over the Amazon, where simulated precipitation peaks around local noon (red colors), compared to late afternoon or early evening in observations (yellow–green).
Diurnal cycles of precipitation in the (a) uniform 1° and (b) VR simulations, as well as (c) TRMM observations. The hue represents the timing of maximum precipitation while saturation denotes the strength of the diurnal cycle. All times are local, with 0000 being midnight and 1200 being noon. Grid transitions in the variable resolution simulation are outlined in black in (b).
Citation: Journal of Climate 28, 7; 10.1175/JCLI-D-14-00599.1
Some improvements are seen regionally in the refined simulation (Fig. 16b) when compared to the uniform 1° run (Fig. 16a). In particular, precipitation occurs later in the day along the South American coast, Central America, southern Mexico, western Africa, and over islands in the Caribbean such as Cuba and Hispaniola. Additionally, a marginal improvement in timing is observed over the southeastern United States. Improvements in the strength of the diurnal cycle can also be seen, primarily along the western coast of South America and up through Central America. The improvements in the VR simulation are likely tied to both improved orography and better representation of high-intensity precipitation rates as discussed in section 5b.
While some regional areas of improvement are noted with increased resolution, refinement does not alleviate the significant large-scale biases that exist in CAM. This is not surprising in light of the results contained in Dirmeyer et al. (2012), who found that resolution only improves the diurnal cycle of precipitation when model grid spacing reaches levels that can support resolved convection (approximately 10 km). Significant improvements at resolutions coarser than these grid spacings (such as those in this study) will likely require modification or reformulation of CAM’s convective subgrid parameterizations.
e. Mountain-gap winds
A mountain-gap wind that passes southward through the Sierra Madre range is prevalent during boreal fall, winter, and spring. This feature is due to an increased pressure gradient between the Gulf of Mexico and the tropical Pacific during cold air outbreaks over North America. The pressure gradient results in a cross-isthmus wind that is funneled into a gap just north of the Gulf of Tehuantepec, creating a low-level jet strong enough to appear in the mean climatology (Chelton et al. 2000). These features are shown to be associated with tropical cyclogenesis in the eastern Pacific (Holbach and Bourassa 2014), and therefore, their representation within the model may be critical to producing realistic storm climatology.
Figure 17 shows the lowest model level January meridional wind (V wind) for the two models (Figs. 17a,b), and observations (Fig. 17c) from the NOAA Blended Sea Winds dataset (Zhang et al. 2006). Topography is again shown in the lower panels (Figs. 17d–f). Note that the scales for Figs. 17a and 17b and Fig. 17c are not the same since the observed winds are surface winds (as opposed to the lowest model level, which lies at approximately 65 m in CAM) and are also a blend of various products that have different effective averaging times. The jet associated with the aforementioned phenomenon is seen as the maximum in southward wind located at approximately 15°N and 95°W. This feature is significantly more robust and localized in the VR simulation (Fig. 17a) when compared to the coarse 1° run (Fig. 17b). It is apparent that the improved topographical representation in the VR simulation (Fig. 17d) leads to improved local dynamics in this region. Better simulations of features such as these will lead to improved local climate representation in areas that are strongly influenced by orographic features at spatial scales below the typical resolution of most global climate models.
Average January meridional wind near the Gulf of Tehuantepec for the lowest model level in the (a) VR and (b) uniform 1° simulations as well as (c) at the surface from SeaWinds observations. Topography for the same regions is shown for (d),(e) both models as well as (f) the NGDC dataset. The transition boundary between 0.25° and 0.5° in the VR grid is highlighted in black and red in (a) and (d), respectively.
Citation: Journal of Climate 28, 7; 10.1175/JCLI-D-14-00599.1
6. Discussion and conclusions
We have presented climatological results comparing a global simulation with a refined nest to an identically forced simulation without the nest. Using CAM-SE with the latest CAM5 physics package, it is found that the addition of a high-resolution refinement over the Atlantic Ocean (approximately one-tenth of the global domain) does not have any noticeable impact on the global circulation. Global averages are well matched to the previously published CAM5 results from Bacmeister et al. (2014) and Wehner et al. (2014).
When just the Atlantic is isolated, an expected increase (decrease) in large-scale (convective) precipitation is observed with the VR mesh. These effects partially compensate each other with regard to the total modeled precipitation, although total precipitation still increases slightly in the high-resolution nest. The only other parameterized variable showing >5% difference in its long-term mean between the refined and unrefined simulations is the sensible heat flux. We hypothesize this may be a manifestation of the increased resolved wind speeds with horizontal refinement, although additional work is required to isolate this mechanism.
Minor statistical differences are seen between the far-field portion of the grid (region over the North Pacific with identical grid spacing). Minor differences (<3%) are apparent in cloud fraction and surface fluxes. It is unclear whether these are physical differences due to the grid or a result of the smaller spatial averaging region (compared to the global domain). Zarzycki and Jablonowski (2014) show the high-resolution nest may provide some upscale effect to the global domain via extratropical transition of better-resolved tropical cyclones, which may have an impact on the downstream atmosphere, although further analysis is necessary to understand the magnitude of these effects.
Features such as equatorial waves and zonal jets over the west coast of Africa are not disturbed by the addition of refinement, implying atmospheric flow passes through transition regions without developing significant errors in phase speed or other wave reflection/distortion. The frequency of precipitation extremes are increased in the high-resolution nest, which agrees with previous research using CAM (Zarzycki et al. 2014b; Wehner et al. 2014).
Variable resolution allows for a better representation of topography in the high-resolution nest. Improvement is seen in quantities that are spatially correlated with topographical signatures. We highlight improvements in the simulation of total precipitable water and precipitation in the vicinity of mountains and mountain-gap winds using VR grids. Other topographically affected flow may be better represented in global models through this framework. However, we emphasize that this was an atmospheric refinement study centered on a specific ocean basin, with relatively small regions of steep orography included in the high-resolution domain. A more rigorous undertaking is required to fully understand the climatological impact of refinement centered over land or mountainous areas. This is a target for future research.
This study shows that, while the overall spatial pattern of precipitation is improved with increased resolution, possible noise may occur in the precipitation field in the presence of the sharper topography within the refined nest. While not extensively investigated, this phenomenon is likely induced by approximate discontinuities in the surface boundary condition (orography) and may be analogous to Gibbs ringing seen in global spectral models near sharp gradients. Preliminary work has shown that increasing the divergent component of the explicit diffusion (hyperviscosity) within the model’s dynamical core improves some of this noise in the vicinity of rougher topography (Lauritzen et al. 2012). In addition, improved smoothing techniques may also provide better results. These are areas of planned testing and development with variable-resolution CAM.
In addition, while the scalar hyperviscosity described in Levy et al. (2013) and Zarzycki et al. (2014a) and used here handles explicit diffusion in transition regions adequately, a new tensor-based hyperviscosity has been shown to improve results of VR grids using shallow-water test cases (Guba et al. 2014). This formulation allows for a more consistent treatment of dissipation in highly distorted elements and will be tested as an update to the treatment of hyperviscosity in forthcoming simulations using CAM. In addition, new techniques for grid generation will provide less distorted transition regions in future variable-resolution CAM-SE applications (Guba et al. 2014).
Acknowledgments
The authors thank Peter Lauritzen for his assistance with topography generation on VR grids, Chris Skinner for providing useful scripts regarding the African wave analysis, and Paul Ullrich for contributing computing time to complete a subset of the simulations. The authors also acknowledge Lucas Harris and two anonymous reviewers for useful comments which improved this manuscript. C.M.Z, C.J., and D.R.T. were supported by the U.S. Department of Energy, Office of Science Awards DE-SC0003990 and DE-SC0006684. M.A.T. was supported by supported by the DOE Office of Biological and Environmental Research, work packages 12-015334 and 11-014996. Some of this work was completed during the “Multiscale Numerics for the Atmosphere and Ocean” program at the Issac Newton Institute for Mathematical Sciences in Cambridge, United Kingdom. TRMM and NCEP Reanalysis data provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, from their website at www.esrl.noaa.gov/psd/. MERRA data was provided by the Global Modeling and Assimilation Office (GMAO) at NASA Goddard Space Flight Center through the NASA GES DISC online archive. Portions of the data analysis were completed using the Community Earth System Model Atmosphere Model Working Group variability package, available at www.cgd.ucar.edu/amp/amwg/vdiag/index.html.
REFERENCES
Bacmeister, J. T., M. F. Wehner, R. B. Neale, A. Gettelman, C. Hannay, P. H. Lauritzen, J. M. Caron, and J. E. Truesdale, 2014: Exploratory high-resolution climate simulations using the Community Atmosphere Model (CAM). J. Climate, 27, 3073–3099, doi:10.1175/JCLI-D-13-00387.1.
Bosilovich, M. G., F. R. Robertson, and J. Chen, 2011: Global energy and water budgets in MERRA. J. Climate, 24, 5721–5739, doi:10.1175/2011JCLI4175.1.
Boyle, J., and S. A. Klein, 2010: Impact of horizontal resolution on climate model forecasts of tropical precipitation and diabatic heating for the TWP-ICE period. J. Geophys. Res., 115, D23113, doi:10.1029/2010JD014262.
Burpee, R. W., 1972: The origin and structure of easterly waves in the lower troposphere of North Africa. J. Atmos. Sci., 29, 77–90, doi:10.1175/1520-0469(1972)029<0077:TOASOE>2.0.CO;2.
Caian, M., and J.-F. Geleyn, 1997: Some limits to the variable-mesh solution and comparison with the nested-LAM solution. Quart. J. Roy. Meteor. Soc., 123, 743–766, doi:10.1002/qj.49712353911.
Chelton, D. B., M. H. Freilich, and S. K. Esbensen, 2000: Satellite observations of the wind jets off the Pacific coast of Central America. Part I: Case studies and statistical characteristics. Mon. Wea. Rev., 128, 1993–2018, doi:10.1175/1520-0493(2000)128<1993:SOOTWJ>2.0.CO;2.
Chen, C.-T., and T. Knutson, 2008: On the verification and comparison of extreme rainfall indices from climate models. J. Climate, 21, 1605–1621, doi:10.1175/2007JCLI1494.1.
Craig, A. P., M. Vertenstein, and R. Jacob, 2012: A new flexible coupler for earth system modeling developed for CCSM4 and CESM1. Int. J. High Perform. Comput. Appl., 26, 31–42, doi:10.1177/1094342011428141.
Dai, A., 2006: Precipitation characteristics in eighteen coupled climate models. J. Climate, 19, 4605–4630, doi:10.1175/JCLI3884.1.
Dennis, J. M., and Coauthors, 2012: CAM-SE: A scalable spectral element dynamical core for the Community Atmosphere Model. Int. J. High Perform. Comput. Appl., 26, 74–89, doi:10.1177/1094342011428142.
Dirmeyer, P. A., and Coauthors, 2012: Simulating the diurnal cycle of rainfall in global climate models: Resolution versus parameterization. Climate Dyn., 39, 399–418, doi:10.1007/s00382-011-1127-9.
Duffy, P., B. Govindasamy, J. Iorio, J. Milovich, K. Sperber, K. Taylor, M. Wehner, and S. Thompson, 2003: High-resolution simulations of global climate, part 1: Present climate. Climate Dyn., 21, 371–390, doi:10.1007/s00382-003-0339-z.
Evans, K. J., P. H. Lauritzen, S. K. Mishra, R. B. Neale, M. A. Taylor, and J. J. Tribbia, 2013: AMIP simulation with the CAM4 spectral element dynamical core. J. Climate, 26, 689–709, doi:10.1175/JCLI-D-11-00448.1.
Flato, G., and Coauthors, 2014: Evaluation of climate models. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 741–866.
Fox-Rabinovitz, M., J. Côté, B. Dugas, M. Déqué, and J. L. McGregor, 2006: Variable resolution general circulation models: Stretched-grid model intercomparison project (SGMIP). J. Geophys. Res.,111, D16104, doi:10.1029/2005JD006520.
Frank, N. L., 1970: Atlantic tropical systems of 1969. Mon. Wea. Rev., 98, 307–314, doi:10.1175/1520-0493(1970)098<0307:ATSO>2.3.CO;2.
Gates, W. L., 1992: AMIP: The Atmospheric Model Intercomparison Project. Bull. Amer. Meteor. Soc., 73, 1962–1970, doi:10.1175/1520-0477(1992)073<1962:ATAMIP>2.0.CO;2.
Gent, P. R., S. G. Yeager, R. B. Neale, S. Levis, and D. A. Bailey, 2010: Improvements in a half degree atmosphere/land version of the CCSM. Climate Dyn., 34, 819–833, doi:10.1007/s00382-009-0614-8.
Guba, O., M. A. Taylor, P. A. Ullrich, J. R. Overfelt, and M. N. Levy, 2014: The spectral element method (SEM) on variable-resolution grids: Evaluating grid sensitivity and resolution-aware numerical viscosity. Geosci. Model Dev., 7, 2803–2816, doi:10.5194/gmd-7-2803-2014.
Hansen, J., R. Ruedy, M. Sato, and K. Lo, 2010: Global surface temperature change. Rev. Geophys., 48, doi:10.1029/2010RG000345.
Harris, L. M., and S.-J. Lin, 2013: A two-way nested global-regional dynamical core on the cubed-sphere grid. Mon. Wea. Rev., 141, 283–306, doi:10.1175/MWR-D-11-00201.1.
Harris, L. M., and S.-J. Lin, 2014: Global-to-regional nested-grid climate simulations in the GFDL high-resolution atmosphere model. J. Climate, 27, 4890–4910, doi:10.1175/JCLI-D-13-00596.1.
Holbach, H. M., and M. A. Bourassa, 2014: The effects of gap-wind-induced vorticity, the monsoon trough, and the ITCZ on east Pacific tropical cyclogenesis. Mon. Wea. Rev., 142, 1312–1325, doi:10.1175/MWR-D-13-00218.1.
Huffman, G. J., and Coauthors, 2007: The TRMM Multisatellite Precipitation Analysis (TMPA): Quasi-global, multiyear, combined-sensor precipitation estimates at fine scales. J. Hydrometeor., 8, 38–55, doi:10.1175/JHM560.1.
Hurrell, J. W., J. J. Hack, D. Shea, J. M. Caron, and J. Rosinski, 2008: A new sea surface temperature and sea ice boundary dataset for the Community Atmosphere Model. J. Climate, 21, 5145–5153, doi:10.1175/2008JCLI2292.1.
Jablonowski, C., R. C. Oehmke, and Q. F. Stout, 2009: Block-structured adaptive meshes and reduced grids for atmospheric general circulation models. Philos. Trans. Royal Soc. London, A367, 4497–4522, doi:10.1098/rsta.2009.0150.
Jiang, J. H., and Coauthors, 2012: Evaluation of cloud and water vapor simulations in CMIP5 climate models using NASA “A-Train” satellite observations. J. Geophys. Res.,117, D14105, doi:10.1029/2011JD017237; Corrigendum, 118, 11087, doi:10.1002/jgrd.50864.
Jones, P. D., M. New, D. E. Parker, S. Martin, and I. G. Rigor, 1999: Surface air temperature and its changes over the past 150 years. Rev. Geophys., 37, 173–199, doi:10.1029/1999RG900002.
Jung, T., and Coauthors, 2012: High-resolution global climate simulations with the ECMWF model in Project Athena: Experimental design, model climate, and seasonal forecast skill. J. Climate, 25, 3155–3172, doi:10.1175/JCLI-D-11-00265.1.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471, doi:10.1175/1520-0477(1996)077<0437:TNYRP>2.0.CO;2.
Kiehl, J. T., T. L. Schneider, P. J. Rasch, M. C. Barth, and J. Wong, 2000: Radiative forcing due to sulfate aerosols from simulations with the National Center for Atmospheric Research Community Climate Model, version 3. J. Geophys. Res., 105, 1441–1457, doi:10.1029/1999JD900495.
Kiladis, G. N., M. C. Wheeler, P. T. Haertel, K. H. Straub, and P. E. Roundy, 2009: Convectively coupled equatorial waves. Rev. Geophys.,47, RG2003, doi:10.1029/2008RG000266.
Laprise, R., and Coauthors, 2008: Challenging some tenets of regional climate modelling. Meteor. Atmos. Phys., 100, 3–22, doi:10.1007/s00703-008-0292-9.
Lauritzen, P. H., J. Bacmeister, M. A. Taylor, and R. B. Neale, 2012: New CAM (NSF-DOE Community Atmosphere Model) topography generation software: CAM5.2. 2012 Fall Meeting, San Francisco, CA, Amer. Geophys. Union, Abstract A53C-0157.
Levy, M., J. Overfelt, and M. A. Taylor, 2013: A variable resolution spectral element dynamical core in the Community Atmosphere Model. Tech. Note 2013-0697J, Sandia National Laboratories, Albuquerque, NM, 25 pp.
Li, F., W. D. Collins, M. F. Wehner, D. L. Williamson, J. G. Olson, and C. Algieri, 2011: Impact of horizontal resolution on simulation of precipitation extremes in an aqua-planet version of Community Atmospheric Model (CAM3). Tellus, 63A, 884–892, doi:10.1111/j.1600-0870.2011.00544.x.
Liu, X., and Coauthors, 2012: Toward a minimal representation of aerosols in climate models: description and evaluation in the Community Atmosphere Model CAM5. Geosci. Model Dev., 5, 709–739, doi:10.5194/gmd-5-709-2012.
Lorant, V., and J. Royer, 2001: Sensitivity of equatorial convection to horizontal resolution in aquaplanet simulations with a variable-resolution GCM. Mon. Wea. Rev., 129, 2730–2745, doi:10.1175/1520-0493(2001)129<2730:SOECTH>2.0.CO;2.
Manganello, J. V., and Coauthors, 2012: Tropical cyclone climatology in a 10-km global atmospheric GCM: Toward weather-resolving climate modeling. J. Climate, 25, 3867–3893, doi:10.1175/JCLI-D-11-00346.1.
McDonald, A., 2003: Transparent boundary conditions for the shallow-water equations: Testing in a nested environment. Mon. Wea. Rev., 131, 698–705, doi:10.1175/1520-0493(2003)131<0698:TBCFTS>2.0.CO;2.
Mehran, A., A. AghaKouchak, and T. J. Phillips, 2014: Evaluation of CMIP5 continental precipitation simulations relative to satellite-based gauge-adjusted observations. J. Geophys. Res. Atmos., 119, 1695–1707, doi:10.1002/2013JD021152.
Mesinger, F., and K. Veljovic, 2013: Limited area NWP and regional climate modeling: A test of the relaxation vs Eta lateral boundary conditions. Meteor. Atmos. Phys., 119, 1–16, doi:10.1007/s00703-012-0217-5.
Mishra, S. K., M. A. Taylor, R. D. Nair, P. H. Lauritzen, H. M. Tufo, and J. J. Tribbia, 2011: Evaluation of the HOMME dynamical core in the aquaplanet configuration of NCAR CAM4: Rainfall. J. Climate, 24, 4037–4055, doi:10.1175/2011JCLI3860.1.
Neale, R. B., and Coauthors, 2010a: Description of the NCAR Community Atmosphere Model (CAM 4.0). NCAR Tech. Note NCAR/TN-485+STR, 212 pp. [Available online at www.cesm.ucar.edu/models/ccsm4.0/cam/docs/description/cam4_desc.pdf.]
Neale, R. B., and Coauthors, 2010b: Description of the NCAR Community Atmosphere Model (CAM5.0). NCAR Tech. Note NCAR/TN-486+STR, 268 pp. [Available online at www.cesm.ucar.edu/models/cesm1.1/cam/docs/description/cam5_desc.pdf.]
Neale, R. B., J. Richter, S. Park, P. H. Lauritzen, S. J. Vavrus, P. J. Rasch, and M. Zhang, 2013: The mean climate of the Community Atmosphere Model (CAM4) in forced SST and fully coupled experiments. J. Climate, 26, 5150–5168, doi:10.1175/JCLI-D-12-00236.1.
O’Brien, T. A., F. Li, W. D. Collins, S. A. Rauscher, T. D. Ringler, M. A. Taylor, S. M. Hagos, and R. L. Leung, 2013: Observed scaling in clouds and precipitation and scale incognizance in regional to global atmospheric models. J. Climate, 26, 9313–9333, doi:10.1175/JCLI-D-13-00005.1.
Ohfuchi, W., and Coauthors, 2004: 10-km mesh meso-scale resolving simulations of the global atmosphere on the earth simulator: Preliminary outcomes of AFES (AGCM for the Earth Simulator). J. Earth Simulator, 1, 8–34.
Rauscher, S. A., T. D. Ringler, W. C. Skamarock, and A. A. Mirin, 2013: Exploring a global multiresolution modeling approach using aquaplanet simulations. J. Climate, 26, 2432–2452, doi:10.1175/JCLI-D-12-00154.1.
Reed, R. J., D. C. Norquist, and E. E. Recker, 1977: The structure and properties of African wave disturbances as observed during Phase III of GATE. Mon. Wea. Rev., 105, 317–333, doi:10.1175/1520-0493(1977)105<0317:TSAPOA>2.0.CO;2.
Richter, J. H., F. Sassi, and R. R. Garcia, 2010: Toward a physically based gravity wave source parameterization in a general circulation model. J. Atmos. Sci., 67, 136–156, doi:10.1175/2009JAS3112.1.
Rienecker, M. M., and Coauthors, 2011: MERRA: NASA’s Modern-Era Retrospective Analysis for Research and applications. J. Climate, 24, 3624–3648, doi:10.1175/JCLI-D-11-00015.1.
Ringler, T., L. Ju, and M. Gunzburger, 2008: A multiresolution method for climate system modeling: Application of spherical centroidal Voronoi tessellations. Ocean Dyn., 58, 475–498, doi:10.1007/s10236-008-0157-2.
Rossow, W. B., and R. A. Schiffer, 1999: Advances in understanding clouds from ISCCP. Bull. Amer. Meteor. Soc., 80, 2261–2287, doi:10.1175/1520-0477(1999)080<2261:AIUCFI>2.0.CO;2.
Rougier, J., D. M. H. Sexton, J. M. Murphy, and D. Stainforth, 2009: Analyzing the climate sensitivity of the HadSM3 climate model using ensembles from different but related experiments. J. Climate, 22, 3540–3557, doi:10.1175/2008JCLI2533.1.
Skamarock, W. C., J. B. Klemp, M. G. Duda, L. D. Fowler, S.-H. Park, and T. D. Ringler, 2012: A multiscale nonhydrostatic atmospheric model using centroidal Voronoi tesselations and C-grid staggering. Mon. Wea. Rev., 140, 3090–3105, doi:10.1175/MWR-D-11-00215.1.
Skinner, C. B., and N. S. Diffenbaugh, 2013: The contribution of African easterly waves to monsoon precipitation in the CMIP3 ensemble. J. Geophys. Res., 118, 3590–3609, doi:10.1002/jgrd.50363.
Taylor, K. E., 2001: Summarizing multiple aspects of model performance in a single diagram. J. Geophys. Res., 106, 7183–7192, doi:10.1029/2000JD900719.
Taylor, M. A., 2011: Conservation of mass and energy for the moist atmospheric primitive equations on unstructured grids. Numerical Techniques for Global Atmospheric Models, P. H. Lauritzen et al., Eds., Lecture Notes in Computational Science and Engineering, Vol. 80, Springer, 357–380.
Taylor, M. A., and A. Fournier, 2010: A compatible and conservative spectral element method on unstructured grids. J. Comput. Phys., 229, 5879–5895, doi:10.1016/j.jcp.2010.04.008.
Taylor, M. A., J. Tribbia, and M. Iskandarani, 1997: The spectral element method for the shallow water equations on the sphere. J. Comput. Phys., 130, 92–108, doi:10.1006/jcph.1996.5554.
Thomas, S., and R. Loft, 2005: The NCAR spectral element climate dynamical core: Semi-implicit Eulerian formulation. J. Sci. Comput., 25, 307–322, doi:10.1007/s10915-004-4646-2.
Tomita, H., 2008: A stretched icosahedral grid by a new grid transformation. J. Meteor. Soc. Japan, 86A, 107–119, doi:10.2151/jmsj.86A.107.
Tuller, S. E., 1968: World distribution of mean monthly and annual precipitable water. Mon. Wea. Rev., 96, 785–797, doi:10.1175/1520-0493(1968)096<0785:WDOMMA>2.0.CO;2.
Walko, R. L., and R. Avissar, 2011: A direct method for constructing refined regions in unstructured conforming triangular–hexagonal computational grids: Application to OLAM. Mon. Wea. Rev., 139, 3923–3937, doi:10.1175/MWR-D-11-00021.1.
Warner, T. T., R. A. Peterson, and R. E. Treadon, 1997: A tutorial on lateral boundary conditions as a basic and potentially serious limitation to regional numerical weather prediction. Bull. Amer. Meteor. Soc., 78, 2599–2617, doi:10.1175/1520-0477(1997)078<2599:ATOLBC>2.0.CO;2.
Wehner, M. F., and Coauthors, 2014: The effect of horizontal resolution on simulation quality in the Community Atmospheric Model, CAM5.1. J. Adv. Model. Earth Syst.,6, 980–997, doi:10.1002/2013MS000276.
Wheeler, M., and G. N. Kiladis, 1999: Convectively coupled equatorial waves: Analysis of clouds and temperature in the wavenumber–frequency domain. J. Atmos. Sci., 56, 374–399, doi:10.1175/1520-0469(1999)056<0374:CCEWAO>2.0.CO;2.
Williamson, D. L., 2008: Convergence of aqua-planet simulations with increasing resolution in the Community Atmospheric Model, version3. Tellus, 60A, 848–862, doi:10.1111/j.1600-0870.2008.00339.x.
Williamson, D. L., 2013: The effect of time steps and time-scales on parametrization suites. Quart. J. Roy. Meteor. Soc., 139, 548–560, doi:10.1002/qj.1992.
Zarzycki, C. M., and C. Jablonowski, 2014: A multidecadal simulation of Atlantic tropical cyclones using a variable-resolution global atmospheric general circulation model. J. Adv. Model. Earth Syst., 6, 805–828, doi:10.1002/2014MS000352.
Zarzycki, C. M., C. Jablonowski, and M. A. Taylor, 2014a: Using variable resolution meshes to model tropical cyclones in the Community Atmosphere Model. Mon. Wea. Rev., 142, 1221–1239, doi:10.1175/MWR-D-13-00179.1.
Zarzycki, C. M., M. N. Levy, C. Jablonowski, J. R. Overfelt, M. A. Taylor, and P. A. Ullrich, 2014b: Aquaplanet experiments using CAM’s variable-resolution dynamical core. J. Climate, 27, 5481–5503, doi:10.1175/JCLI-D-14-00004.1.
Zhang, H.-M., J. J. Bates, and R. W. Reynolds, 2006: Assessment of composite global sampling: Sea surface wind speed. Geophys. Res. Lett., 33, L17714, doi:10.1029/2006GL027086.
Zhao, M., I. M. Held, S.-J. Lin, and G. A. Vecchi, 2009: Simulations of global hurricane climatology, interannual variability, and response to global warming using a 50-km resolution GCM. J. Climate, 22, 6653–6678, doi:10.1175/2009JCLI3049.1.
Zhao, M., I. M. Held, and S.-J. Lin, 2012: Some counterintuitive dependencies of tropical cyclone frequency on parameters in a GCM. J. Atmos. Sci., 69, 2272–2283, doi:10.1175/JAS-D-11-0238.1.
